Google (NASDAQ: GOOGL) has discontinued the image-generating feature of its flagship chatbot, Gemini, after attempts to encourage diversity resulted in “embarrassing” and “offensive” images.
Google launched the new feature three weeks ago to compete with market leaders such as Midjourney, Stable Diffusion, and OpenAI’s DALL-E 3. But last week the feature made headlines for controversial results, including its inclusion of Native Americans and Asians. Create “Portraits of America’s Founding Fathers.”
According to Google AI, America’s Founding Fathers, Vikings, and Popes are: pic.twitter.com/lw4aIKLwkp
— EndWokeness (@EndWokeness) February 21, 2024
In response, Google discontinued the feature with Senior Vice President Prabhakar Raghavan saying:
“It’s clear this feature misses the mark,” he admitted.
According to Raghavan, Google wanted Gemini to work for everyone around the world. Likewise, when we asked for images of people walking their dogs, we included diversity in our development so that we only received images of white people.
But Gemini had two problems. First, Google failed to account for cases in which an image should be out of scope, such as when requesting a white teacher or historically accurate event.
Second, Google created Gemini too cautiously because it was too concerned about it falling into the pitfalls faced by other AI models, such as producing racist or sexually violent results. Several users stated that Gemini would refuse to generate images that could be used by other AI models.
We are already working to resolve recent issues with Gemini’s image creation capabilities. In the meantime, we are temporarily suspending portrait image creation and will soon re-release an improved version. https://t.co/SLxYPGoqOZ
— Google Communications (@Google_Comms) February 22, 2024
Last August, Margaret Mitchell, chief ethics scientist at Hugging Face, an AI startup valued at $4.5 billion, speculated that the problem may be that Google added diversity language “inside” Gemini to make it more inclusive.
Mitchell, who previously served as head of ethical AI at Google, says that when a user types in a command like “portrait of a chief,” Google’s LLM adds the term “indigenous” to get better results.
Mitchell also speculated that Google may have a problem with prioritizing more diverse results higher.
“Rather than focusing on these after-the-fact solutions, we need to focus on the data. “If we manage our data well from the beginning, we don’t need to have racist systems,” she told the Washington Post.
Google has promised to continue fine-tuning its image generator.
“I can’t promise that Gemini won’t sometimes produce results that are embarrassing, inaccurate, or offensive,” says Raghavan. “But I can promise you that we will continue to take action whenever issues are identified.”
For artificial intelligence (AI) to function properly within the law and succeed in the face of growing challenges, it must integrate enterprise blockchain systems that ensure data input quality and ownership. This helps keep your data safe while ensuring immutability. data. Check out CoinGeek’s coverage Learn more about this new technology. Why enterprise blockchain will become the backbone of AI.
See: AI and blockchain
Are you new to blockchain? To learn more about blockchain technology, check out CoinGeek’s Blockchain for Beginners section, our ultimate resource guide.